Abstract: Aiming at the problem of low accuracy of wind speed prediction in wind farms,an ultra-short term wind speed prediction method based on wind speed fluctuation feature extraction is proposed. Firstly,the joint probability density model of wind speed and wind speed variation is established to analyze the uncertainty characteristics of wind speed. According to the wind speed fluctuation characteristics,the wind speed decomposition is recombined into fluctuation and trend quantity by using the values of Ensemble Empirical Mode Decomposition（EEMD） and Sample Entropy（SampEn） of wind speed components. The Artificial Fish Swarm Algorithm（AFSA） is used to optimize the Wavelet Neural Network（WNN） for trend prediction. The Improved Non-linear Autoregression（INARX）with external input neural network is used to predict the wind speed fluctuation,and then the predicted wind speed is obtained. Through the actual wind farm wind speed simulation prediction and comparison with many forecasting methods,the results show that the prediction error of this method is low,and can accurately predict the ultra-short term wind speed.